Perplexity and NVIDIA Researchers Introduce JOLT: a Joint Teacher-Student Reinforcement Learning Method
Perplexity, along with NVIDIA team, has introduced JOLT, a new post-training method for language models.
Proof1 independent outlet
Key points
- JOLT stands for Joint On-policy Learning and Teaching.
- On MATH, JOLT matched GRPO’s accuracy with 6.5x fewer processed training tokens.
- What it is: a post-training method.
- JOLT is a self-distillation method from Perplexity and NVIDIA that trains one model as teacher and student.
Sources (1)
- [1]Perplexity and NVIDIA Researchers Introduce JOLT: a Joint Teacher-Student Reinforcement Learning MethodMarkTechPost · Oct 22, 04:56 AM
Perplexity, along with NVIDIA team, has introduced JOLT, a new post-training method for language models.
JOLT stands for Joint On-policy Learning and Teaching.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 21, 2026Recursive shows first result of its AI-driven research process with Flint
- Oct 21, 2026Mistral Advances AI for Engineering with NVIDIA
- Oct 20, 2026Fine-tuning GPT-OSS-20B for multilingual reasoning on Amazon SageMaker HyperPod
- Oct 20, 2026Voxtral 3
- Oct 20, 2026[AINews] DeepSeek DeepInfer: Open Source SOTA Inference running 155T toks/day/cluster making $22m/day at ~80% margin for a 81 day payback
- Oct 20, 2026Self-Evolving Agents through Privileged Retrospective Distillation
